Sarim-Hash/Qwen3-14B-sandbagging Overview

This model is a fine-tuned iteration of the Qwen3-14B base model, developed by Sarim-Hash. It features 14 billion parameters and was trained with a context length of 32768 tokens.

Training Details

The model underwent fine-tuning on the df_final dataset. Key training hyperparameters include:

• Learning Rate: 1e-05
• Batch Size: 1 (train), 8 (eval)
• Gradient Accumulation Steps: 2
• Optimizer: AdamW with betas=(0.9, 0.999) and epsilon=1e-08
• LR Scheduler: Cosine type with a warmup ratio of 0.1
• Epochs: 8.0
• Distributed Training: Multi-GPU setup across 4 devices

Current Status

As of now, specific details regarding the model's intended uses, limitations, and detailed performance evaluation are not yet available in the provided documentation. Users are encouraged to consult future updates for more comprehensive information on its capabilities and optimal applications.